diff --git a/source/Lib/CommonLib/Buffer.cpp b/source/Lib/CommonLib/Buffer.cpp
index 2e9c1a42a3935f1e992b83663e57b2b895965168..2454c501ae976e3885395f6efcc3169531db8c9b 100644
--- a/source/Lib/CommonLib/Buffer.cpp
+++ b/source/Lib/CommonLib/Buffer.cpp
@@ -63,7 +63,7 @@ void addAvgCore( const T* src1, int src1Stride, const T* src2, int src2Stride, T
}
#if JVET_L0256_BIO
-void addBIOAvgCore(const Pel* src0, int src0Stride, const Pel* src1, int src1Stride, Pel *dst, int dstStride, const Pel *pGradX0, const Pel *pGradX1, const Pel *pGradY0, const Pel*pGradY1, int gradStride, int width, int height, int tmpx, int tmpy, int shift, int offset, const ClpRng& clpRng)
+void addBIOAvgCore(const Pel* src0, int src0Stride, const Pel* src1, int src1Stride, Pel *dst, int dstStride, const Pel *gradX0, const Pel *gradX1, const Pel *gradY0, const Pel*gradY1, int gradStride, int width, int height, int tmpx, int tmpy, int shift, int offset, const ClpRng& clpRng)
{
int b = 0;
@@ -71,125 +71,125 @@ void addBIOAvgCore(const Pel* src0, int src0Stride, const Pel* src1, int src1Str
{
for (int x = 0; x < width; x += 4)
{
- b = tmpx * (pGradX0[x] - pGradX1[x]) + tmpy * (pGradY0[x] - pGradY1[x]);
+ b = tmpx * (gradX0[x] - gradX1[x]) + tmpy * (gradY0[x] - gradY1[x]);
b = ((b + 1) >> 1);
dst[x] = ClipPel((int16_t)rightShift((src0[x] + src1[x] + b + offset), shift), clpRng);
- b = tmpx * (pGradX0[x + 1] - pGradX1[x + 1]) + tmpy * (pGradY0[x + 1] - pGradY1[x + 1]);
+ b = tmpx * (gradX0[x + 1] - gradX1[x + 1]) + tmpy * (gradY0[x + 1] - gradY1[x + 1]);
b = ((b + 1) >> 1);
dst[x + 1] = ClipPel((int16_t)rightShift((src0[x + 1] + src1[x + 1] + b + offset), shift), clpRng);
- b = tmpx * (pGradX0[x + 2] - pGradX1[x + 2]) + tmpy * (pGradY0[x + 2] - pGradY1[x + 2]);
+ b = tmpx * (gradX0[x + 2] - gradX1[x + 2]) + tmpy * (gradY0[x + 2] - gradY1[x + 2]);
b = ((b + 1) >> 1);
dst[x + 2] = ClipPel((int16_t)rightShift((src0[x + 2] + src1[x + 2] + b + offset), shift), clpRng);
- b = tmpx * (pGradX0[x + 3] - pGradX1[x + 3]) + tmpy * (pGradY0[x + 3] - pGradY1[x + 3]);
+ b = tmpx * (gradX0[x + 3] - gradX1[x + 3]) + tmpy * (gradY0[x + 3] - gradY1[x + 3]);
b = ((b + 1) >> 1);
dst[x + 3] = ClipPel((int16_t)rightShift((src0[x + 3] + src1[x + 3] + b + offset), shift), clpRng);
}
dst += dstStride; src0 += src0Stride; src1 += src1Stride;
- pGradX0 += gradStride; pGradX1 += gradStride; pGradY0 += gradStride; pGradY1 += gradStride;
+ gradX0 += gradStride; gradX1 += gradStride; gradY0 += gradStride; gradY1 += gradStride;
}
}
-void gradFilterCore(Pel* pSrc, int srcStride, int width, int height, int gradStride, Pel* pGradX, Pel* pGradY)
+void gradFilterCore(Pel* pSrc, int srcStride, int width, int height, int gradStride, Pel* gradX, Pel* gradY)
{
- Pel* piSrcTmp = pSrc + srcStride + 1;
- Pel* piGradXTmp = pGradX + gradStride + 1;
- Pel* piGradYTmp = pGradY + gradStride + 1;
+ Pel* srcTmp = pSrc + srcStride + 1;
+ Pel* gradXTmp = gradX + gradStride + 1;
+ Pel* gradYTmp = gradY + gradStride + 1;
- for (int y = 0; y < (height - 2 * JVET_L0256_BIO_EXTEND_SIZE); y++)
+ for (int y = 0; y < (height - 2 * BIO_EXTEND_SIZE); y++)
{
- for (int x = 0; x < (width - 2 * JVET_L0256_BIO_EXTEND_SIZE); x++)
+ for (int x = 0; x < (width - 2 * BIO_EXTEND_SIZE); x++)
{
- piGradYTmp[x] = (piSrcTmp[x + srcStride] - piSrcTmp[x - srcStride]) >> 4;
- piGradXTmp[x] = (piSrcTmp[x + 1] - piSrcTmp[x - 1]) >> 4;
+ gradYTmp[x] = (srcTmp[x + srcStride] - srcTmp[x - srcStride]) >> 4;
+ gradXTmp[x] = (srcTmp[x + 1] - srcTmp[x - 1]) >> 4;
}
- piGradXTmp += gradStride;
- piGradYTmp += gradStride;
- piSrcTmp += srcStride;
+ gradXTmp += gradStride;
+ gradYTmp += gradStride;
+ srcTmp += srcStride;
}
- piGradXTmp = pGradX + gradStride + 1;
- piGradYTmp = pGradY + gradStride + 1;
- for (int y = 0; y < (height - 2 * JVET_L0256_BIO_EXTEND_SIZE); y++)
+ gradXTmp = gradX + gradStride + 1;
+ gradYTmp = gradY + gradStride + 1;
+ for (int y = 0; y < (height - 2 * BIO_EXTEND_SIZE); y++)
{
- piGradXTmp[-1] = piGradXTmp[0];
- piGradXTmp[width - 2 * JVET_L0256_BIO_EXTEND_SIZE] = piGradXTmp[width - 2 * JVET_L0256_BIO_EXTEND_SIZE - 1];
- piGradXTmp += gradStride;
+ gradXTmp[-1] = gradXTmp[0];
+ gradXTmp[width - 2 * BIO_EXTEND_SIZE] = gradXTmp[width - 2 * BIO_EXTEND_SIZE - 1];
+ gradXTmp += gradStride;
- piGradYTmp[-1] = piGradYTmp[0];
- piGradYTmp[width - 2 * JVET_L0256_BIO_EXTEND_SIZE] = piGradYTmp[width - 2 * JVET_L0256_BIO_EXTEND_SIZE - 1];
- piGradYTmp += gradStride;
+ gradYTmp[-1] = gradYTmp[0];
+ gradYTmp[width - 2 * BIO_EXTEND_SIZE] = gradYTmp[width - 2 * BIO_EXTEND_SIZE - 1];
+ gradYTmp += gradStride;
}
- piGradXTmp = pGradX + gradStride;
- piGradYTmp = pGradY + gradStride;
- ::memcpy(piGradXTmp - gradStride, piGradXTmp, sizeof(Pel)*(width));
- ::memcpy(piGradXTmp + (height - 2 * JVET_L0256_BIO_EXTEND_SIZE)*gradStride, piGradXTmp + (height - 2 * JVET_L0256_BIO_EXTEND_SIZE - 1)*gradStride, sizeof(Pel)*(width));
- ::memcpy(piGradYTmp - gradStride, piGradYTmp, sizeof(Pel)*(width));
- ::memcpy(piGradYTmp + (height - 2 * JVET_L0256_BIO_EXTEND_SIZE)*gradStride, piGradYTmp + (height - 2 * JVET_L0256_BIO_EXTEND_SIZE - 1)*gradStride, sizeof(Pel)*(width));
+ gradXTmp = gradX + gradStride;
+ gradYTmp = gradY + gradStride;
+ ::memcpy(gradXTmp - gradStride, gradXTmp, sizeof(Pel)*(width));
+ ::memcpy(gradXTmp + (height - 2 * BIO_EXTEND_SIZE)*gradStride, gradXTmp + (height - 2 * BIO_EXTEND_SIZE - 1)*gradStride, sizeof(Pel)*(width));
+ ::memcpy(gradYTmp - gradStride, gradYTmp, sizeof(Pel)*(width));
+ ::memcpy(gradYTmp + (height - 2 * BIO_EXTEND_SIZE)*gradStride, gradYTmp + (height - 2 * BIO_EXTEND_SIZE - 1)*gradStride, sizeof(Pel)*(width));
}
-void calcBIOParCore(const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGradX0, const Pel* pGradX1, const Pel* pGradY0, const Pel* pGradY1, int* m_piDotProductTemp1, int* m_piDotProductTemp2, int* m_piDotProductTemp3, int* m_piDotProductTemp5, int* m_piDotProductTemp6, const int iSrc0Stride, const int iSrc1Stride, const int iGradStride, const int iWidthG, const int iHeightG)
+void calcBIOParCore(const Pel* srcY0Temp, const Pel* srcY1Temp, const Pel* gradX0, const Pel* gradX1, const Pel* gradY0, const Pel* gradY1, int* dotProductTemp1, int* dotProductTemp2, int* dotProductTemp3, int* dotProductTemp5, int* dotProductTemp6, const int src0Stride, const int src1Stride, const int gradStride, const int widthG, const int heightG)
{
- for (int y = 0; y < iHeightG; y++)
+ for (int y = 0; y < heightG; y++)
{
- for (int x = 0; x < iWidthG; x++)
+ for (int x = 0; x < widthG; x++)
{
- int temp = (pSrcY0Temp[x] >> 6) - (pSrcY1Temp[x] >> 6);
- int tempX = (pGradX0[x] + pGradX1[x]) >> 3;
- int tempY = (pGradY0[x] + pGradY1[x]) >> 3;
- m_piDotProductTemp1[x] = tempX * tempX;
- m_piDotProductTemp2[x] = tempX * tempY;
- m_piDotProductTemp3[x] = -tempX * temp;
- m_piDotProductTemp5[x] = tempY * tempY;
- m_piDotProductTemp6[x] = -tempY * temp;
+ int temp = (srcY0Temp[x] >> 6) - (srcY1Temp[x] >> 6);
+ int tempX = (gradX0[x] + gradX1[x]) >> 3;
+ int tempY = (gradY0[x] + gradY1[x]) >> 3;
+ dotProductTemp1[x] = tempX * tempX;
+ dotProductTemp2[x] = tempX * tempY;
+ dotProductTemp3[x] = -tempX * temp;
+ dotProductTemp5[x] = tempY * tempY;
+ dotProductTemp6[x] = -tempY * temp;
}
- pSrcY0Temp += iSrc0Stride;
- pSrcY1Temp += iSrc1Stride;
- pGradX0 += iGradStride;
- pGradX1 += iGradStride;
- pGradY0 += iGradStride;
- pGradY1 += iGradStride;
- m_piDotProductTemp1 += iWidthG;
- m_piDotProductTemp2 += iWidthG;
- m_piDotProductTemp3 += iWidthG;
- m_piDotProductTemp5 += iWidthG;
- m_piDotProductTemp6 += iWidthG;
+ srcY0Temp += src0Stride;
+ srcY1Temp += src1Stride;
+ gradX0 += gradStride;
+ gradX1 += gradStride;
+ gradY0 += gradStride;
+ gradY1 += gradStride;
+ dotProductTemp1 += widthG;
+ dotProductTemp2 += widthG;
+ dotProductTemp3 += widthG;
+ dotProductTemp5 += widthG;
+ dotProductTemp6 += widthG;
}
}
void calcBlkGradientCore(int sx, int sy, int *arraysGx2, int *arraysGxGy, int *arraysGxdI, int *arraysGy2, int *arraysGydI, int &sGx2, int &sGy2, int &sGxGy, int &sGxdI, int &sGydI, int width, int height, int unitSize)
{
- int *pGx2 = arraysGx2;
- int *pGy2 = arraysGy2;
- int *pGxGy = arraysGxGy;
- int *pGxdI = arraysGxdI;
- int *pGydI = arraysGydI;
+ int *Gx2 = arraysGx2;
+ int *Gy2 = arraysGy2;
+ int *GxGy = arraysGxGy;
+ int *GxdI = arraysGxdI;
+ int *GydI = arraysGydI;
// set to the above row due to JVET_K0485_BIO_EXTEND_SIZE
- pGx2 -= (JVET_L0256_BIO_EXTEND_SIZE*width);
- pGy2 -= (JVET_L0256_BIO_EXTEND_SIZE*width);
- pGxGy -= (JVET_L0256_BIO_EXTEND_SIZE*width);
- pGxdI -= (JVET_L0256_BIO_EXTEND_SIZE*width);
- pGydI -= (JVET_L0256_BIO_EXTEND_SIZE*width);
+ Gx2 -= (BIO_EXTEND_SIZE*width);
+ Gy2 -= (BIO_EXTEND_SIZE*width);
+ GxGy -= (BIO_EXTEND_SIZE*width);
+ GxdI -= (BIO_EXTEND_SIZE*width);
+ GydI -= (BIO_EXTEND_SIZE*width);
- for (int y = -JVET_L0256_BIO_EXTEND_SIZE; y < unitSize + JVET_L0256_BIO_EXTEND_SIZE; y++)
+ for (int y = -BIO_EXTEND_SIZE; y < unitSize + BIO_EXTEND_SIZE; y++)
{
- for (int x = -JVET_L0256_BIO_EXTEND_SIZE; x < unitSize + JVET_L0256_BIO_EXTEND_SIZE; x++)
+ for (int x = -BIO_EXTEND_SIZE; x < unitSize + BIO_EXTEND_SIZE; x++)
{
- sGx2 += pGx2[x];
- sGy2 += pGy2[x];
- sGxGy += pGxGy[x];
- sGxdI += pGxdI[x];
- sGydI += pGydI[x];
+ sGx2 += Gx2[x];
+ sGy2 += Gy2[x];
+ sGxGy += GxGy[x];
+ sGxdI += GxdI[x];
+ sGydI += GydI[x];
}
- pGx2 += width;
- pGy2 += width;
- pGxGy += width;
- pGxdI += width;
- pGydI += width;
+ Gx2 += width;
+ Gy2 += width;
+ GxGy += width;
+ GxdI += width;
+ GydI += width;
}
}
#endif
diff --git a/source/Lib/CommonLib/Buffer.h b/source/Lib/CommonLib/Buffer.h
index b8e69315f295aac1b1c4732b06dae3bd5e2e8c5a..a0142743f3efd1b287efa3c039b80e038aeb6db6 100644
--- a/source/Lib/CommonLib/Buffer.h
+++ b/source/Lib/CommonLib/Buffer.h
@@ -69,9 +69,9 @@ struct PelBufferOps
void ( *linTf4 ) ( const Pel* src0, int src0Stride, Pel *dst, int dstStride, int width, int height, int scale, int shift, int offset, const ClpRng& clpRng, bool bClip );
void ( *linTf8 ) ( const Pel* src0, int src0Stride, Pel *dst, int dstStride, int width, int height, int scale, int shift, int offset, const ClpRng& clpRng, bool bClip );
#if JVET_L0256_BIO
- void(*addBIOAvg4) (const Pel* src0, int src0Stride, const Pel* src1, int src1Stride, Pel *dst, int dstStride, const Pel *pGradX0, const Pel *pGradX1, const Pel *pGradY0, const Pel*pGradY1, int gradStride, int width, int height, int tmpx, int tmpy, int shift, int offset, const ClpRng& clpRng);
- void(*bioGradFilter) (Pel* pSrc, int srcStride, int width, int height, int gradStride, Pel* pGradX, Pel* pGradY);
- void(*calcBIOPar) (const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGradX0, const Pel* pGradX1, const Pel* pGradY0, const Pel* pGradY1, int* m_piDotProductTemp1, int* m_piDotProductTemp2, int* m_piDotProductTemp3, int* m_piDotProductTemp5, int* m_piDotProductTemp6, const int iSrc0Stride, const int iSrc1Stride, const int iGradStride, const int iWidthG, const int iHeightG);
+ void(*addBIOAvg4) (const Pel* src0, int src0Stride, const Pel* src1, int src1Stride, Pel *dst, int dstStride, const Pel *gradX0, const Pel *gradX1, const Pel *gradY0, const Pel*gradY1, int gradStride, int width, int height, int tmpx, int tmpy, int shift, int offset, const ClpRng& clpRng);
+ void(*bioGradFilter) (Pel* pSrc, int srcStride, int width, int height, int gradStride, Pel* gradX, Pel* gradY);
+ void(*calcBIOPar) (const Pel* srcY0Temp, const Pel* srcY1Temp, const Pel* gradX0, const Pel* gradX1, const Pel* gradY0, const Pel* gradY1, int* dotProductTemp1, int* dotProductTemp2, int* dotProductTemp3, int* dotProductTemp5, int* dotProductTemp6, const int src0Stride, const int src1Stride, const int gradStride, const int widthG, const int heightG);
void(*calcBlkGradient)(int sx, int sy, int *arraysGx2, int *arraysGxGy, int *arraysGxdI, int *arraysGy2, int *arraysGydI, int &sGx2, int &sGy2, int &sGxGy, int &sGxdI, int &sGydI, int width, int height, int unitSize);
#endif
#if ENABLE_SIMD_OPT_GBI
diff --git a/source/Lib/CommonLib/CommonDef.h b/source/Lib/CommonLib/CommonDef.h
index 98174dbeb777a52e254f4a21c4897c03b7cc0f68..41d8be8c326c47a95fdbd4bf4792a44a5f2bd618 100644
--- a/source/Lib/CommonLib/CommonDef.h
+++ b/source/Lib/CommonLib/CommonDef.h
@@ -297,6 +297,12 @@ static const int MAX_NUM_GT2_BINS_4x4SUBBLOCK = 4; ///< max
static const int MAX_NUM_REG_BINS_2x2SUBBLOCK = 8; ///< max number of context-coded bins (incl. gt2 bins) per 2x2 subblock (chroma)
static const int MAX_NUM_GT2_BINS_2x2SUBBLOCK = 2; ///< max number of gt2 bins per 2x2 subblock (chroma)
#endif
+
+#if JVET_L0256_BIO
+static const int BIO_EXTEND_SIZE = 1;
+static const int BIO_TEMP_BUFFER_SIZE = (MAX_CU_SIZE + 2 * BIO_EXTEND_SIZE) * (MAX_CU_SIZE + 2 * BIO_EXTEND_SIZE);
+#endif
+
#if JVET_L0646_GBI
static const int GBI_NUM = 5; ///< the number of weight options
static const int GBI_DEFAULT = ((uint8_t)(GBI_NUM >> 1)); ///< Default weighting index representing for w=0.5
diff --git a/source/Lib/CommonLib/InterPrediction.cpp b/source/Lib/CommonLib/InterPrediction.cpp
index 4720c41efb5bcff76a85033f6ab5bca045bcc612..d980b24cc05c874241e46be5acb47342d03830ad 100644
--- a/source/Lib/CommonLib/InterPrediction.cpp
+++ b/source/Lib/CommonLib/InterPrediction.cpp
@@ -56,10 +56,10 @@ InterPrediction::InterPrediction()
, m_maxCompIDToPred ( MAX_NUM_COMPONENT )
, m_pcRdCost ( nullptr )
#if JVET_L0256_BIO
-, m_pGradX0(nullptr)
-, m_pGradY0(nullptr)
-, m_pGradX1(nullptr)
-, m_pGradY1(nullptr)
+, m_gradX0(nullptr)
+, m_gradY0(nullptr)
+, m_gradX1(nullptr)
+, m_gradY1(nullptr)
, m_subPuMC(false)
#endif
{
@@ -118,10 +118,10 @@ void InterPrediction::destroy()
}
#if JVET_L0256_BIO
- xFree(m_pGradX0); m_pGradX0 = nullptr;
- xFree(m_pGradY0); m_pGradY0 = nullptr;
- xFree(m_pGradX1); m_pGradX1 = nullptr;
- xFree(m_pGradY1); m_pGradY1 = nullptr;
+ xFree(m_gradX0); m_gradX0 = nullptr;
+ xFree(m_gradY0); m_gradY0 = nullptr;
+ xFree(m_gradX1); m_gradX1 = nullptr;
+ xFree(m_gradY1); m_gradY1 = nullptr;
#endif
}
@@ -142,8 +142,8 @@ void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC )
for( uint32_t c = 0; c < MAX_NUM_COMPONENT; c++ )
{
#if JVET_L0256_BIO
- int extWidth = MAX_CU_SIZE + (2 * JVET_L0256_BIO_EXTEND_SIZE + 2) + 16;
- int extHeight = MAX_CU_SIZE + (2 * JVET_L0256_BIO_EXTEND_SIZE + 2) + 1;
+ int extWidth = MAX_CU_SIZE + (2 * BIO_EXTEND_SIZE + 2) + 16;
+ int extHeight = MAX_CU_SIZE + (2 * BIO_EXTEND_SIZE + 2) + 1;
#else
int extWidth = MAX_CU_SIZE + 16;
int extHeight = MAX_CU_SIZE + 1;
@@ -169,10 +169,10 @@ void InterPrediction::init( RdCost* pcRdCost, ChromaFormat chromaFormatIDC )
m_iRefListIdx = -1;
#if JVET_L0256_BIO
- m_pGradX0 = (Pel*)xMalloc(Pel, BIO_TEMP_BUFFER_SIZE);
- m_pGradY0 = (Pel*)xMalloc(Pel, BIO_TEMP_BUFFER_SIZE);
- m_pGradX1 = (Pel*)xMalloc(Pel, BIO_TEMP_BUFFER_SIZE);
- m_pGradY1 = (Pel*)xMalloc(Pel, BIO_TEMP_BUFFER_SIZE);
+ m_gradX0 = (Pel*)xMalloc(Pel, BIO_TEMP_BUFFER_SIZE);
+ m_gradY0 = (Pel*)xMalloc(Pel, BIO_TEMP_BUFFER_SIZE);
+ m_gradX1 = (Pel*)xMalloc(Pel, BIO_TEMP_BUFFER_SIZE);
+ m_gradY1 = (Pel*)xMalloc(Pel, BIO_TEMP_BUFFER_SIZE);
#endif
}
@@ -336,7 +336,7 @@ void InterPrediction::xSubPuMC( PredictionUnit& pu, PelUnitBuf& predBuf, const R
void InterPrediction::xPredInterUni(const PredictionUnit& pu, const RefPicList& eRefPicList, PelUnitBuf& pcYuvPred, const bool& bi
#if JVET_L0256_BIO
- ,const bool& bBIOApplied /*=false*/
+ ,const bool& bioApplied /*=false*/
#endif
)
{
@@ -368,7 +368,7 @@ void InterPrediction::xPredInterUni(const PredictionUnit& pu, const RefPicList&
if ( pu.cu->affine )
{
#if JVET_L0256_BIO
- CHECK( bBIOApplied, "BIO is not allowed with affine" );
+ CHECK( bioApplied, "BIO is not allowed with affine" );
#endif
xPredAffineBlk( compID, pu, pu.cu->slice->getRefPic( eRefPicList, iRefIdx ), mv, pcYuvPred, bi, pu.cu->slice->clpRng( compID ) );
}
@@ -376,7 +376,7 @@ void InterPrediction::xPredInterUni(const PredictionUnit& pu, const RefPicList&
{
xPredInterBlk( compID, pu, pu.cu->slice->getRefPic( eRefPicList, iRefIdx ), mv[0], pcYuvPred, bi, pu.cu->slice->clpRng( compID )
#if JVET_L0256_BIO
- ,bBIOApplied
+ ,bioApplied
#endif
);
}
@@ -389,31 +389,31 @@ void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred)
const Slice &slice = *pu.cs->slice;
#if JVET_L0256_BIO
- bool bBIOApplied = false;
+ bool bioApplied = false;
if (pu.cs->sps->getSpsNext().getUseBIO())
{
if (pu.cu->affine || m_subPuMC)
{
- bBIOApplied = false;
+ bioApplied = false;
}
else
{
- const bool bBIOcheck0 = !(pps.getWPBiPred() && slice.getSliceType() == B_SLICE);
- const bool bBIOcheck1 = !(pps.getUseWP() && slice.getSliceType() == P_SLICE);
- if (bBIOcheck0
- && bBIOcheck1
+ const bool biocheck0 = !(pps.getWPBiPred() && slice.getSliceType() == B_SLICE);
+ const bool biocheck1 = !(pps.getUseWP() && slice.getSliceType() == P_SLICE);
+ if (biocheck0
+ && biocheck1
&& PU::isBiPredFromDifferentDir(pu)
&& !(pu.Y().height == 4 || (pu.Y().width == 4 && pu.Y().height == 8))
)
{
- bBIOApplied = true;
+ bioApplied = true;
}
}
#if JVET_L0646_GBI
- if (pu.cu->cs->sps->getSpsNext().getUseGBi() && bBIOApplied && pu.cu->GBiIdx != GBI_DEFAULT)
+ if (pu.cu->cs->sps->getSpsNext().getUseGBi() && bioApplied && pu.cu->GBiIdx != GBI_DEFAULT)
{
- bBIOApplied = false;
+ bioApplied = false;
}
#endif
}
@@ -439,7 +439,7 @@ void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred)
{
xPredInterUni ( pu, eRefPicList, pcMbBuf, true
#if JVET_L0256_BIO
- ,bBIOApplied
+ ,bioApplied
#endif
);
}
@@ -474,7 +474,7 @@ void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred)
else
{
#if JVET_L0256_BIO
- xWeightedAverage( pu, srcPred0, srcPred1, pcYuvPred, slice.getSPS()->getBitDepths(), slice.clpRngs(), bBIOApplied );
+ xWeightedAverage( pu, srcPred0, srcPred1, pcYuvPred, slice.getSPS()->getBitDepths(), slice.clpRngs(), bioApplied );
#else
xWeightedAverage( pu, srcPred0, srcPred1, pcYuvPred, slice.getSPS()->getBitDepths(), slice.clpRngs() );
#endif
@@ -483,7 +483,7 @@ void InterPrediction::xPredInterBi(PredictionUnit& pu, PelUnitBuf &pcYuvPred)
void InterPrediction::xPredInterBlk ( const ComponentID& compID, const PredictionUnit& pu, const Picture* refPic, const Mv& _mv, PelUnitBuf& dstPic, const bool& bi, const ClpRng& clpRng
#if JVET_L0256_BIO
- ,const bool& bBIOApplied /*=false*/
+ ,const bool& bioApplied /*=false*/
#endif
)
{
@@ -533,10 +533,10 @@ void InterPrediction::xPredInterBlk ( const ComponentID& compID, const Predictio
Pel *backupDstBufPtr = dstBuf.buf;
int backupDstBufStride = dstBuf.stride;
- if (bBIOApplied && compID == COMPONENT_Y)
+ if (bioApplied && compID == COMPONENT_Y)
{
- width = width + 2 * JVET_L0256_BIO_EXTEND_SIZE + 2;
- height = height + 2 * JVET_L0256_BIO_EXTEND_SIZE + 2;
+ width = width + 2 * BIO_EXTEND_SIZE + 2;
+ height = height + 2 * BIO_EXTEND_SIZE + 2;
// change MC output
dstBuf.stride = width;
@@ -582,7 +582,7 @@ void InterPrediction::xPredInterBlk ( const ComponentID& compID, const Predictio
JVET_J0090_SET_CACHE_ENABLE( true );
}
#if JVET_L0256_BIO
- if (bBIOApplied && compID == COMPONENT_Y)
+ if (bioApplied && compID == COMPONENT_Y)
{
refBuf.buf = refBuf.buf - refBuf.stride - 1;
dstBuf.buf = m_filteredBlockTmp[2 + m_iRefListIdx][compID] + dstBuf.stride + 1;
@@ -744,48 +744,48 @@ int getMSB( unsigned x )
}
#if JVET_L0256_BIO
-void InterPrediction::applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf &pcYuvSrc0, const CPelUnitBuf &pcYuvSrc1, const int &iRefIdx0, const int &iRefIdx1, PelUnitBuf &pcYuvDst, const BitDepths &clipBitDepths)
+void InterPrediction::applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf &yuvSrc0, const CPelUnitBuf &yuvSrc1, const int &refIdx0, const int &refIdx1, PelUnitBuf &yuvDst, const BitDepths &clipBitDepths)
{
- const int iHeight = pcYuvDst.Y().height;
- const int iWidth = pcYuvDst.Y().width;
- int iHeightG = iHeight + 2 * JVET_L0256_BIO_EXTEND_SIZE;
- int iWidthG = iWidth + 2 * JVET_L0256_BIO_EXTEND_SIZE;
- int offsetPos = iWidthG*JVET_L0256_BIO_EXTEND_SIZE + JVET_L0256_BIO_EXTEND_SIZE;
-
- Pel* pGradX0 = m_pGradX0;
- Pel* pGradX1 = m_pGradX1;
- Pel* pGradY0 = m_pGradY0;
- Pel* pGradY1 = m_pGradY1;
-
- int stridePredMC = iWidthG + 2;
- const Pel* pSrcY0 = m_filteredBlockTmp[2][COMPONENT_Y] + stridePredMC + 1;
- const Pel* pSrcY1 = m_filteredBlockTmp[3][COMPONENT_Y] + stridePredMC + 1;
- const int iSrc0Stride = stridePredMC;
- const int iSrc1Stride = stridePredMC;
-
- Pel* pDstY = pcYuvDst.Y().buf;
- const int iDstStride = pcYuvDst.Y().stride;
- const Pel* pSrcY0Temp = pSrcY0;
- const Pel* pSrcY1Temp = pSrcY1;
+ const int height = yuvDst.Y().height;
+ const int width = yuvDst.Y().width;
+ int heightG = height + 2 * BIO_EXTEND_SIZE;
+ int widthG = width + 2 * BIO_EXTEND_SIZE;
+ int offsetPos = widthG*BIO_EXTEND_SIZE + BIO_EXTEND_SIZE;
+
+ Pel* gradX0 = m_gradX0;
+ Pel* gradX1 = m_gradX1;
+ Pel* gradY0 = m_gradY0;
+ Pel* gradY1 = m_gradY1;
+
+ int stridePredMC = widthG + 2;
+ const Pel* srcY0 = m_filteredBlockTmp[2][COMPONENT_Y] + stridePredMC + 1;
+ const Pel* srcY1 = m_filteredBlockTmp[3][COMPONENT_Y] + stridePredMC + 1;
+ const int src0Stride = stridePredMC;
+ const int src1Stride = stridePredMC;
+
+ Pel* dstY = yuvDst.Y().buf;
+ const int dstStride = yuvDst.Y().stride;
+ const Pel* srcY0Temp = srcY0;
+ const Pel* srcY1Temp = srcY1;
for (int refList = 0; refList < NUM_REF_PIC_LIST_01; refList++)
{
Pel* dstTempPtr = m_filteredBlockTmp[2 + refList][COMPONENT_Y] + stridePredMC + 1;
- Pel* pGradY = (refList == 0) ? m_pGradY0 : m_pGradY1;
- Pel* pGradX = (refList == 0) ? m_pGradX0 : m_pGradX1;
+ Pel* gradY = (refList == 0) ? m_gradY0 : m_gradY1;
+ Pel* gradX = (refList == 0) ? m_gradX0 : m_gradX1;
- g_pelBufOP.bioGradFilter(dstTempPtr, stridePredMC, iWidthG, iHeightG, iWidthG, pGradX, pGradY);
- Pel* pcPadStr = m_filteredBlockTmp[2 + refList][COMPONENT_Y] + 2 * stridePredMC + 2;
- for (int y = 0; y< iHeight; y++)
+ g_pelBufOP.bioGradFilter(dstTempPtr, stridePredMC, widthG, heightG, widthG, gradX, gradY);
+ Pel* padStr = m_filteredBlockTmp[2 + refList][COMPONENT_Y] + 2 * stridePredMC + 2;
+ for (int y = 0; y< height; y++)
{
- pcPadStr[-1] = pcPadStr[0];
- pcPadStr[iWidth] = pcPadStr[iWidth - 1];
- pcPadStr += stridePredMC;
+ padStr[-1] = padStr[0];
+ padStr[width] = padStr[width - 1];
+ padStr += stridePredMC;
}
- pcPadStr = m_filteredBlockTmp[2 + refList][COMPONENT_Y] + 2 * stridePredMC + 1;
- ::memcpy(pcPadStr - stridePredMC, pcPadStr, sizeof(Pel)*(iWidthG));
- ::memcpy(pcPadStr + iHeight*stridePredMC, pcPadStr + (iHeight - 1)*stridePredMC, sizeof(Pel)*(iWidthG));
+ padStr = m_filteredBlockTmp[2 + refList][COMPONENT_Y] + 2 * stridePredMC + 1;
+ ::memcpy(padStr - stridePredMC, padStr, sizeof(Pel)*(widthG));
+ ::memcpy(padStr + height*stridePredMC, padStr + (height - 1)*stridePredMC, sizeof(Pel)*(widthG));
}
const ClpRng& clpRng = pu.cu->cs->slice->clpRng(COMPONENT_Y);
@@ -794,20 +794,20 @@ void InterPrediction::applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf
const int offset = (1 << (shiftNum - 1)) + 2 * IF_INTERNAL_OFFS;
const int limit = ((int)1 << (4 + IF_INTERNAL_PREC - bitDepth - 5));
- int* m_piDotProductTemp1 = m_piDotProduct1;
- int* m_piDotProductTemp2 = m_piDotProduct2;
- int* m_piDotProductTemp3 = m_piDotProduct3;
- int* m_piDotProductTemp5 = m_piDotProduct5;
- int* m_piDotProductTemp6 = m_piDotProduct6;
+ int* dotProductTemp1 = m_dotProduct1;
+ int* dotProductTemp2 = m_dotProduct2;
+ int* dotProductTemp3 = m_dotProduct3;
+ int* dotProductTemp5 = m_dotProduct5;
+ int* dotProductTemp6 = m_dotProduct6;
- g_pelBufOP.calcBIOPar(pSrcY0Temp, pSrcY1Temp, pGradX0, pGradX1, pGradY0, pGradY1, m_piDotProductTemp1, m_piDotProductTemp2, m_piDotProductTemp3, m_piDotProductTemp5, m_piDotProductTemp6, iSrc0Stride, iSrc1Stride, iWidthG, iWidthG, iHeightG);
+ g_pelBufOP.calcBIOPar(srcY0Temp, srcY1Temp, gradX0, gradX1, gradY0, gradY1, dotProductTemp1, dotProductTemp2, dotProductTemp3, dotProductTemp5, dotProductTemp6, src0Stride, src1Stride, widthG, widthG, heightG);
- int xUnit = (iWidth >> 2);
- int yUnit = (iHeight >> 2);
+ int xUnit = (width >> 2);
+ int yUnit = (height >> 2);
- Pel *pDstY0 = pDstY;
- pGradX0 = m_pGradX0; pGradX1 = m_pGradX1;
- pGradY0 = m_pGradY0; pGradY1 = m_pGradY1;
+ Pel *dstY0 = dstY;
+ gradX0 = m_gradX0; gradX1 = m_gradX1;
+ gradY0 = m_gradY0; gradY1 = m_gradY1;
for (int yu = 0; yu < yUnit; yu++)
{
@@ -815,23 +815,23 @@ void InterPrediction::applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf
{
if (m_bioPredSubBlkDist[yu*xUnit + xu] < m_bioSubBlkDistThres)
{
- pSrcY0Temp = pSrcY0 + (stridePredMC + 1) + ((yu*iSrc0Stride + xu) << 2);
- pSrcY1Temp = pSrcY1 + (stridePredMC + 1) + ((yu*iSrc1Stride + xu) << 2);
- pDstY0 = pDstY + ((yu*iDstStride + xu) << 2);
- g_pelBufOP.addAvg4(pSrcY0Temp, iSrc0Stride, pSrcY1Temp, iSrc1Stride, pDstY0, iDstStride, (1 << 2), (1 << 2), shiftNum, offset, clpRng);
+ srcY0Temp = srcY0 + (stridePredMC + 1) + ((yu*src0Stride + xu) << 2);
+ srcY1Temp = srcY1 + (stridePredMC + 1) + ((yu*src1Stride + xu) << 2);
+ dstY0 = dstY + ((yu*dstStride + xu) << 2);
+ g_pelBufOP.addAvg4(srcY0Temp, src0Stride, srcY1Temp, src1Stride, dstY0, dstStride, (1 << 2), (1 << 2), shiftNum, offset, clpRng);
continue;
}
int sGxdI = 0, sGydI = 0, sGxGy = 0, sGx2 = 0, sGy2 = 0;
int tmpx = 0, tmpy = 0;
- m_piDotProductTemp1 = m_piDotProduct1 + offsetPos + ((yu*iWidthG + xu) << 2);
- m_piDotProductTemp2 = m_piDotProduct2 + offsetPos + ((yu*iWidthG + xu) << 2);
- m_piDotProductTemp3 = m_piDotProduct3 + offsetPos + ((yu*iWidthG + xu) << 2);
- m_piDotProductTemp5 = m_piDotProduct5 + offsetPos + ((yu*iWidthG + xu) << 2);
- m_piDotProductTemp6 = m_piDotProduct6 + offsetPos + ((yu*iWidthG + xu) << 2);
+ dotProductTemp1 = m_dotProduct1 + offsetPos + ((yu*widthG + xu) << 2);
+ dotProductTemp2 = m_dotProduct2 + offsetPos + ((yu*widthG + xu) << 2);
+ dotProductTemp3 = m_dotProduct3 + offsetPos + ((yu*widthG + xu) << 2);
+ dotProductTemp5 = m_dotProduct5 + offsetPos + ((yu*widthG + xu) << 2);
+ dotProductTemp6 = m_dotProduct6 + offsetPos + ((yu*widthG + xu) << 2);
- g_pelBufOP.calcBlkGradient(xu << 2, yu << 2, m_piDotProductTemp1, m_piDotProductTemp2, m_piDotProductTemp3, m_piDotProductTemp5, m_piDotProductTemp6, sGx2, sGy2, sGxGy, sGxdI, sGydI, iWidthG, iHeightG, (1 << 2));
+ g_pelBufOP.calcBlkGradient(xu << 2, yu << 2, dotProductTemp1, dotProductTemp2, dotProductTemp3, dotProductTemp5, dotProductTemp6, sGx2, sGy2, sGxGy, sGxdI, sGydI, widthG, heightG, (1 << 2));
if (sGx2 > 0)
{
@@ -848,15 +848,15 @@ void InterPrediction::applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf
tmpy = Clip3(-limit, limit, tmpy);
}
- pSrcY0Temp = pSrcY0 + (stridePredMC + 1) + ((yu*iSrc0Stride + xu) << 2);
- pSrcY1Temp = pSrcY1 + (stridePredMC + 1) + ((yu*iSrc0Stride + xu) << 2);
- pGradX0 = m_pGradX0 + offsetPos + ((yu*iWidthG + xu) << 2);
- pGradX1 = m_pGradX1 + offsetPos + ((yu*iWidthG + xu) << 2);
- pGradY0 = m_pGradY0 + offsetPos + ((yu*iWidthG + xu) << 2);
- pGradY1 = m_pGradY1 + offsetPos + ((yu*iWidthG + xu) << 2);
+ srcY0Temp = srcY0 + (stridePredMC + 1) + ((yu*src0Stride + xu) << 2);
+ srcY1Temp = srcY1 + (stridePredMC + 1) + ((yu*src0Stride + xu) << 2);
+ gradX0 = m_gradX0 + offsetPos + ((yu*widthG + xu) << 2);
+ gradX1 = m_gradX1 + offsetPos + ((yu*widthG + xu) << 2);
+ gradY0 = m_gradY0 + offsetPos + ((yu*widthG + xu) << 2);
+ gradY1 = m_gradY1 + offsetPos + ((yu*widthG + xu) << 2);
- pDstY0 = pDstY + ((yu*iDstStride + xu) << 2);
- g_pelBufOP.addBIOAvg4(pSrcY0Temp, iSrc0Stride, pSrcY1Temp, iSrc1Stride, pDstY0, iDstStride, pGradX0, pGradX1, pGradY0, pGradY1, iWidthG, (1 << 2), (1 << 2), (int)tmpx, (int)tmpy, shiftNum, offset, clpRng);
+ dstY0 = dstY + ((yu*dstStride + xu) << 2);
+ g_pelBufOP.addBIOAvg4(srcY0Temp, src0Stride, srcY1Temp, src1Stride, dstY0, dstStride, gradX0, gradX1, gradY0, gradY1, widthG, (1 << 2), (1 << 2), (int)tmpx, (int)tmpy, shiftNum, offset, clpRng);
} // xu
} // yu
}
@@ -957,7 +957,7 @@ bool InterPrediction::xCalcBiPredSubBlkDist(const PredictionUnit &pu, const Pel*
#endif
#if JVET_L0256_BIO
-void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bBIOApplied )
+void InterPrediction::xWeightedAverage(const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bioApplied )
#else
void InterPrediction::xWeightedAverage( const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs )
#endif
@@ -971,17 +971,17 @@ void InterPrediction::xWeightedAverage( const PredictionUnit& pu, const CPelUnit
if( pu.cu->GBiIdx != GBI_DEFAULT )
{
#if JVET_L0256_BIO
- CHECK(bBIOApplied, "GBi is disallowed with BIO");
+ CHECK(bioApplied, "GBi is disallowed with BIO");
#endif
pcYuvDst.addWeightedAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, pu.cu->GBiIdx);
return;
}
#endif
#if JVET_L0256_BIO
- if (bBIOApplied)
+ if (bioApplied)
{
- const int src0Stride = pu.lwidth() + 2 * JVET_L0256_BIO_EXTEND_SIZE + 2;
- const int src1Stride = pu.lwidth() + 2 * JVET_L0256_BIO_EXTEND_SIZE + 2;
+ const int src0Stride = pu.lwidth() + 2 * BIO_EXTEND_SIZE + 2;
+ const int src1Stride = pu.lwidth() + 2 * BIO_EXTEND_SIZE + 2;
const Pel* pSrcY0 = m_filteredBlockTmp[2][COMPONENT_Y] + 2 * src0Stride + 2;
const Pel* pSrcY1 = m_filteredBlockTmp[3][COMPONENT_Y] + 2 * src1Stride + 2;
@@ -995,7 +995,7 @@ void InterPrediction::xWeightedAverage( const PredictionUnit& pu, const CPelUnit
pcYuvDst.bufs[0].addAvg(CPelBuf(pSrcY0, src0Stride, pu.lumaSize()), CPelBuf(pSrcY1, src1Stride, pu.lumaSize()), clpRngs.comp[0]);
}
}
- pcYuvDst.addAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, bBIOApplied);
+ pcYuvDst.addAvg(pcYuvSrc0, pcYuvSrc1, clpRngs, bioApplied);
#else
pcYuvDst.addAvg( pcYuvSrc0, pcYuvSrc1, clpRngs );
#endif
diff --git a/source/Lib/CommonLib/InterPrediction.h b/source/Lib/CommonLib/InterPrediction.h
index 01b13c84e527ecac4a179f20a389041dd64dba22..0bff85b70a5f525b26f8734c6e3b2c682cf48846 100644
--- a/source/Lib/CommonLib/InterPrediction.h
+++ b/source/Lib/CommonLib/InterPrediction.h
@@ -60,10 +60,6 @@ class Mv;
// Class definition
// ====================================================================================================================
-#if JVET_L0256_BIO
-#define BIO_TEMP_BUFFER_SIZE ( MAX_CU_SIZE+2*JVET_L0256_BIO_EXTEND_SIZE ) * ( MAX_CU_SIZE+2*JVET_L0256_BIO_EXTEND_SIZE )
-#endif
-
class InterPrediction : public WeightPrediction
{
private:
@@ -73,11 +69,11 @@ private:
Distortion m_bioSubBlkDistThres;
Distortion m_bioPredSubBlkDist[MAX_NUM_PARTS_IN_CTU];
- int m_piDotProduct1[BIO_TEMP_BUFFER_SIZE];
- int m_piDotProduct2[BIO_TEMP_BUFFER_SIZE];
- int m_piDotProduct3[BIO_TEMP_BUFFER_SIZE];
- int m_piDotProduct5[BIO_TEMP_BUFFER_SIZE];
- int m_piDotProduct6[BIO_TEMP_BUFFER_SIZE];
+ int m_dotProduct1[BIO_TEMP_BUFFER_SIZE];
+ int m_dotProduct2[BIO_TEMP_BUFFER_SIZE];
+ int m_dotProduct3[BIO_TEMP_BUFFER_SIZE];
+ int m_dotProduct5[BIO_TEMP_BUFFER_SIZE];
+ int m_dotProduct6[BIO_TEMP_BUFFER_SIZE];
#endif
protected:
@@ -97,32 +93,32 @@ protected:
int m_iRefListIdx;
#if JVET_L0256_BIO
- Pel* m_pGradX0;
- Pel* m_pGradY0;
- Pel* m_pGradX1;
- Pel* m_pGradY1;
+ Pel* m_gradX0;
+ Pel* m_gradY0;
+ Pel* m_gradX1;
+ Pel* m_gradY1;
bool m_subPuMC;
int rightShiftMSB(int numer, int denom);
- void applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf &pcYuvSrc0, const CPelUnitBuf &pcYuvSrc1, const int &iRefIdx0, const int &iRefIdx1, PelUnitBuf &pcYuvDst, const BitDepths &clipBitDepths);
- bool xCalcBiPredSubBlkDist(const PredictionUnit &pu, const Pel* pYuvSrc0, const int src0Stride, const Pel* pYuvSrc1, const int src1Stride, const BitDepths &clipBitDepths);
+ void applyBiOptFlow(const PredictionUnit &pu, const CPelUnitBuf &yuvSrc0, const CPelUnitBuf &yuvSrc1, const int &refIdx0, const int &refIdx1, PelUnitBuf &yuvDst, const BitDepths &clipBitDepths);
+ bool xCalcBiPredSubBlkDist(const PredictionUnit &pu, const Pel* yuvSrc0, const int src0Stride, const Pel* yuvSrc1, const int src1Stride, const BitDepths &clipBitDepths);
void bioSampleExtendBilinearFilter(Pel const* src, int srcStride, Pel *dst, int dstStride, int width, int height, int dim, int fracX, int fracY, bool isLast, const ChromaFormat fmt, const ClpRng& clpRng);
#endif
void xPredInterUni ( const PredictionUnit& pu, const RefPicList& eRefPicList, PelUnitBuf& pcYuvPred, const bool& bi
#if JVET_L0256_BIO
- ,const bool& bBIOApplied = false
+ ,const bool& bioApplied = false
#endif
);
void xPredInterBi ( PredictionUnit& pu, PelUnitBuf &pcYuvPred );
void xPredInterBlk ( const ComponentID& compID, const PredictionUnit& pu, const Picture* refPic, const Mv& _mv, PelUnitBuf& dstPic, const bool& bi, const ClpRng& clpRng
#if JVET_L0256_BIO
- ,const bool& bBIOApplied = false
+ ,const bool& bioApplied = false
#endif
);
#if JVET_L0256_BIO
- void xWeightedAverage ( const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bBIOApplied );
+ void xWeightedAverage ( const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs, const bool& bioApplied );
#else
void xWeightedAverage ( const PredictionUnit& pu, const CPelUnitBuf& pcYuvSrc0, const CPelUnitBuf& pcYuvSrc1, PelUnitBuf& pcYuvDst, const BitDepths& clipBitDepths, const ClpRngs& clpRngs );
#endif
diff --git a/source/Lib/CommonLib/RdCost.cpp b/source/Lib/CommonLib/RdCost.cpp
index 85306088a19032ef2405bd1742c08198b66f32dc..572eacb8ba30219900a0137a0d55f05157511430 100644
--- a/source/Lib/CommonLib/RdCost.cpp
+++ b/source/Lib/CommonLib/RdCost.cpp
@@ -323,7 +323,7 @@ void RdCost::setDistParam( DistParam &rcDP, const CPelBuf &org, const CPelBuf &c
}
#if JVET_L0256_BIO
-void RdCost::setDistParam( DistParam &rcDP, const Pel* pOrg, const Pel* piRefY, int iOrgStride, int iRefStride, int bitDepth, ComponentID compID, int width, int height, int subShiftMode, int step, bool useHadamard, bool bBIOApplied )
+void RdCost::setDistParam( DistParam &rcDP, const Pel* pOrg, const Pel* piRefY, int iOrgStride, int iRefStride, int bitDepth, ComponentID compID, int width, int height, int subShiftMode, int step, bool useHadamard, bool bioApplied )
#else
void RdCost::setDistParam( DistParam &rcDP, const Pel* pOrg, const Pel* piRefY, int iOrgStride, int iRefStride, int bitDepth, ComponentID compID, int width, int height, int subShiftMode, int step, bool useHadamard )
#endif
@@ -348,7 +348,7 @@ void RdCost::setDistParam( DistParam &rcDP, const Pel* pOrg, const Pel* piRefY,
CHECK( useHadamard || rcDP.useMR || subShiftMode > 0, "only used in xDirectMCCost with these default parameters (so far...)" );
#if JVET_L0256_BIO
- if ( bBIOApplied )
+ if ( bioApplied )
{
rcDP.distFunc = m_afpDistortFunc[ DF_SAD_INTERMEDIATE_BITDEPTH ];
return;
diff --git a/source/Lib/CommonLib/RdCost.h b/source/Lib/CommonLib/RdCost.h
index 07b9f69395c7e622cb6499acf201c4d16603df79..3c95f90adf54158c5cfce66572b7dedc82613f7d 100644
--- a/source/Lib/CommonLib/RdCost.h
+++ b/source/Lib/CommonLib/RdCost.h
@@ -102,9 +102,6 @@ private:
// for distortion
static FpDistFunc m_afpDistortFunc[DF_TOTAL_FUNCTIONS]; // [eDFunc]
-#if JVET_L0256_BIO
-
-#endif
CostMode m_costMode;
double m_distortionWeight[MAX_NUM_COMPONENT]; // only chroma values are used.
double m_dLambda;
@@ -158,7 +155,7 @@ public:
void setDistParam( DistParam &rcDP, const CPelBuf &org, const Pel* piRefY , int iRefStride, int bitDepth, ComponentID compID, int subShiftMode = 0, int step = 1, bool useHadamard = false );
void setDistParam( DistParam &rcDP, const CPelBuf &org, const CPelBuf &cur, int bitDepth, ComponentID compID, bool useHadamard = false );
#if JVET_L0256_BIO
- void setDistParam( DistParam &rcDP, const Pel* pOrg, const Pel* piRefY, int iOrgStride, int iRefStride, int bitDepth, ComponentID compID, int width, int height, int subShiftMode = 0, int step = 1, bool useHadamard = false, bool bBIOApplied = false );
+ void setDistParam( DistParam &rcDP, const Pel* pOrg, const Pel* piRefY, int iOrgStride, int iRefStride, int bitDepth, ComponentID compID, int width, int height, int subShiftMode = 0, int step = 1, bool useHadamard = false, bool bioApplied = false );
#else
void setDistParam( DistParam &rcDP, const Pel* pOrg, const Pel* piRefY, int iOrgStride, int iRefStride, int bitDepth, ComponentID compID, int width, int height, int subShiftMode = 0, int step = 1, bool useHadamard = false );
#endif
@@ -273,7 +270,7 @@ private:
static Distortion xGetSAD_SIMD ( const DistParam& pcDtParam );
template< int iWidth, X86_VEXT vext >
static Distortion xGetSAD_NxN_SIMD( const DistParam& pcDtParam );
-#if JVET_L0256_BIO
+#if ENABLE_SIMD_OPT_BIO
template< X86_VEXT vext >
static Distortion xGetSAD_IBD_SIMD(const DistParam& pcDtParam);
#endif
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index af2cf490b7a86a5545535951e2bb5a30a7ca446b..e6ad40144ce200256e7f02b07bc582741f44c798 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -76,9 +76,6 @@
#define L0074_SUBBLOCK_DEBLOCKING 1
#define JVET_L0256_BIO 1
-#if JVET_L0256_BIO
-#define JVET_L0256_BIO_EXTEND_SIZE 1
-#endif
#define JVET_L0646_GBI 1 // Generalized bi-prediction (GBi)
@@ -258,6 +255,7 @@
#if ENABLE_SIMD_OPT_BUFFER
#define ENABLE_SIMD_OPT_GBI 1 ///< SIMD optimization for GBi
#endif
+#define ENABLE_SIMD_OPT_BIO ( JVET_L0256_BIO && ENABLE_SIMD_OPT ) ///< SIMD optimization for BIO
// End of SIMD optimizations
diff --git a/source/Lib/CommonLib/x86/BufferX86.h b/source/Lib/CommonLib/x86/BufferX86.h
index 1add0996f854645da46a4f205a1c0e6076883f76..92f52655a47940016d708dadc68a152a00a1315b 100644
--- a/source/Lib/CommonLib/x86/BufferX86.h
+++ b/source/Lib/CommonLib/x86/BufferX86.h
@@ -128,9 +128,9 @@ void addAvg_SSE( const int16_t* src0, int src0Stride, const int16_t* src1, int s
}
}
-#if JVET_L0256_BIO
+#if ENABLE_SIMD_OPT_BIO
template< X86_VEXT vext >
-void addBIOAvg4_SSE(const Pel* src0, int src0Stride, const Pel* src1, int src1Stride, Pel *dst, int dstStride, const Pel *pGradX0, const Pel *pGradX1, const Pel *pGradY0, const Pel*pGradY1, int iGradStride, int width, int height, int tmpx, int tmpy, int shift, int offset, const ClpRng& clpRng)
+void addBIOAvg4_SSE(const Pel* src0, int src0Stride, const Pel* src1, int src1Stride, Pel *dst, int dstStride, const Pel *gradX0, const Pel *gradX1, const Pel *gradY0, const Pel*gradY1, int gradStride, int width, int height, int tmpx, int tmpy, int shift, int offset, const ClpRng& clpRng)
{
__m128i mm_tmpx = _mm_unpacklo_epi64(_mm_set1_epi16(tmpx), _mm_set1_epi16(tmpy));
__m128i mm_boffset = _mm_set1_epi32(1);
@@ -142,8 +142,8 @@ void addBIOAvg4_SSE(const Pel* src0, int src0Stride, const Pel* src1, int src1St
{
for (int x = 0; x < width; x += 4)
{
- __m128i mm_a = _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)(pGradX0 + x)), _mm_loadl_epi64((const __m128i *)(pGradY0 + x)));
- __m128i mm_b = _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)(pGradX1 + x)), _mm_loadl_epi64((const __m128i *)(pGradY1 + x)));
+ __m128i mm_a = _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)(gradX0 + x)), _mm_loadl_epi64((const __m128i *)(gradY0 + x)));
+ __m128i mm_b = _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)(gradX1 + x)), _mm_loadl_epi64((const __m128i *)(gradY1 + x)));
mm_a = _mm_sub_epi16(mm_a, mm_b);
mm_b = _mm_mulhi_epi16(mm_a, mm_tmpx);
mm_a = _mm_mullo_epi16(mm_a, mm_tmpx);
@@ -158,82 +158,82 @@ void addBIOAvg4_SSE(const Pel* src0, int src0Stride, const Pel* src1, int src1St
_mm_storel_epi64((__m128i *)(dst + x), mm_sum);
}
dst += dstStride; src0 += src0Stride; src1 += src1Stride;
- pGradX0 += iGradStride; pGradX1 += iGradStride; pGradY0 += iGradStride; pGradY1 += iGradStride;
+ gradX0 += gradStride; gradX1 += gradStride; gradY0 += gradStride; gradY1 += gradStride;
}
}
template< X86_VEXT vext >
-void gradFilter_SSE(Pel* piSrc, int iSrcStride, int iWidth, int iHeight, int iGradStride, Pel* piGradX, Pel* piGradY)
+void gradFilter_SSE(Pel* src, int srcStride, int width, int height, int gradStride, Pel* gradX, Pel* gradY)
{
__m128i vzero = _mm_setzero_si128();
- Pel* piSrcTmp = piSrc + iSrcStride + 1;
- Pel* piGradXTmp = piGradX + iGradStride + 1;
- Pel* piGradYTmp = piGradY + iGradStride + 1;
+ Pel* srcTmp = src + srcStride + 1;
+ Pel* gradXTmp = gradX + gradStride + 1;
+ Pel* gradYTmp = gradY + gradStride + 1;
- int iWidthInside = iWidth - 2 * JVET_L0256_BIO_EXTEND_SIZE;
- int iHeightInside = iHeight - 2 * JVET_L0256_BIO_EXTEND_SIZE;
+ int widthInside = width - 2 * BIO_EXTEND_SIZE;
+ int heightInside = height - 2 * BIO_EXTEND_SIZE;
- assert((iWidthInside & 3) == 0);
+ assert((widthInside & 3) == 0);
- for (int y = 0; y < iHeightInside; y++)
+ for (int y = 0; y < heightInside; y++)
{
int x = 0;
- for (; x < iWidthInside; x += 4)
+ for (; x < widthInside; x += 4)
{
- __m128i mmPixTop = _mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i*)(piSrcTmp + x - iSrcStride)));
- __m128i mmPixBottom = _mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i*)(piSrcTmp + x + iSrcStride)));
- __m128i mmPixLeft = _mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i*)(piSrcTmp + x - 1)));
- __m128i mmPixRight = _mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i*)(piSrcTmp + x + 1)));
+ __m128i mmPixTop = _mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i*)(srcTmp + x - srcStride)));
+ __m128i mmPixBottom = _mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i*)(srcTmp + x + srcStride)));
+ __m128i mmPixLeft = _mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i*)(srcTmp + x - 1)));
+ __m128i mmPixRight = _mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i*)(srcTmp + x + 1)));
__m128i mmGradVer = _mm_srai_epi32(_mm_sub_epi32(mmPixBottom, mmPixTop), 4);
__m128i mmGradHor = _mm_srai_epi32(_mm_sub_epi32(mmPixRight, mmPixLeft), 4);
mmGradVer = _mm_packs_epi32(mmGradVer, vzero);
mmGradHor = _mm_packs_epi32(mmGradHor, vzero);
- _mm_storel_epi64((__m128i *)(piGradYTmp + x), mmGradVer);
- _mm_storel_epi64((__m128i *)(piGradXTmp + x), mmGradHor);
+ _mm_storel_epi64((__m128i *)(gradYTmp + x), mmGradVer);
+ _mm_storel_epi64((__m128i *)(gradXTmp + x), mmGradHor);
}
- piGradXTmp += iGradStride;
- piGradYTmp += iGradStride;
- piSrcTmp += iSrcStride;
+ gradXTmp += gradStride;
+ gradYTmp += gradStride;
+ srcTmp += srcStride;
}
- piGradXTmp = piGradX + iGradStride + 1;
- piGradYTmp = piGradY + iGradStride + 1;
- for (int y = 0; y < iHeightInside; y++)
+ gradXTmp = gradX + gradStride + 1;
+ gradYTmp = gradY + gradStride + 1;
+ for (int y = 0; y < heightInside; y++)
{
- piGradXTmp[-1] = piGradXTmp[0];
- piGradXTmp[iWidthInside] = piGradXTmp[iWidthInside - 1];
- piGradXTmp += iGradStride;
+ gradXTmp[-1] = gradXTmp[0];
+ gradXTmp[widthInside] = gradXTmp[widthInside - 1];
+ gradXTmp += gradStride;
- piGradYTmp[-1] = piGradYTmp[0];
- piGradYTmp[iWidthInside] = piGradYTmp[iWidthInside - 1];
- piGradYTmp += iGradStride;
+ gradYTmp[-1] = gradYTmp[0];
+ gradYTmp[widthInside] = gradYTmp[widthInside - 1];
+ gradYTmp += gradStride;
}
- piGradXTmp = piGradX + iGradStride;
- piGradYTmp = piGradY + iGradStride;
- ::memcpy(piGradXTmp - iGradStride, piGradXTmp, sizeof(Pel)*(iWidth));
- ::memcpy(piGradXTmp + iHeightInside*iGradStride, piGradXTmp + (iHeightInside - 1)*iGradStride, sizeof(Pel)*(iWidth));
- ::memcpy(piGradYTmp - iGradStride, piGradYTmp, sizeof(Pel)*(iWidth));
- ::memcpy(piGradYTmp + iHeightInside*iGradStride, piGradYTmp + (iHeightInside - 1)*iGradStride, sizeof(Pel)*(iWidth));
+ gradXTmp = gradX + gradStride;
+ gradYTmp = gradY + gradStride;
+ ::memcpy(gradXTmp - gradStride, gradXTmp, sizeof(Pel)*(width));
+ ::memcpy(gradXTmp + heightInside*gradStride, gradXTmp + (heightInside - 1)*gradStride, sizeof(Pel)*(width));
+ ::memcpy(gradYTmp - gradStride, gradYTmp, sizeof(Pel)*(width));
+ ::memcpy(gradYTmp + heightInside*gradStride, gradYTmp + (heightInside - 1)*gradStride, sizeof(Pel)*(width));
}
template< X86_VEXT vext >
-void calcBIOPar_SSE(const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGradX0, const Pel* pGradX1, const Pel* pGradY0, const Pel* pGradY1, int* m_piDotProductTemp1, int* m_piDotProductTemp2, int* m_piDotProductTemp3, int* m_piDotProductTemp5, int* m_piDotProductTemp6, const int iSrc0Stride, const int iSrc1Stride, const int iGradStride, const int iWidthG, const int iHeightG)
+void calcBIOPar_SSE(const Pel* srcY0Temp, const Pel* srcY1Temp, const Pel* gradX0, const Pel* gradX1, const Pel* gradY0, const Pel* gradY1, int* dotProductTemp1, int* dotProductTemp2, int* dotProductTemp3, int* dotProductTemp5, int* dotProductTemp6, const int src0Stride, const int src1Stride, const int gradStride, const int widthG, const int heightG)
{
- for (int y = 0; y < iHeightG; y++)
+ for (int y = 0; y < heightG; y++)
{
int x = 0;
- for (; x < ((iWidthG >> 3) << 3); x += 8)
+ for (; x < ((widthG >> 3) << 3); x += 8)
{
- __m128i mmSrcY0Temp = _mm_srai_epi16(_mm_loadu_si128((__m128i*)(pSrcY0Temp + x)), 6);
- __m128i mmSrcY1Temp = _mm_srai_epi16(_mm_loadu_si128((__m128i*)(pSrcY1Temp + x)), 6);
- __m128i mmGradX0 = _mm_loadu_si128((__m128i*)(pGradX0 + x));
- __m128i mmGradX1 = _mm_loadu_si128((__m128i*)(pGradX1 + x));
- __m128i mmGradY0 = _mm_loadu_si128((__m128i*)(pGradY0 + x));
- __m128i mmGradY1 = _mm_loadu_si128((__m128i*)(pGradY1 + x));
+ __m128i mmSrcY0Temp = _mm_srai_epi16(_mm_loadu_si128((__m128i*)(srcY0Temp + x)), 6);
+ __m128i mmSrcY1Temp = _mm_srai_epi16(_mm_loadu_si128((__m128i*)(srcY1Temp + x)), 6);
+ __m128i mmGradX0 = _mm_loadu_si128((__m128i*)(gradX0 + x));
+ __m128i mmGradX1 = _mm_loadu_si128((__m128i*)(gradX1 + x));
+ __m128i mmGradY0 = _mm_loadu_si128((__m128i*)(gradY0 + x));
+ __m128i mmGradY1 = _mm_loadu_si128((__m128i*)(gradY1 + x));
__m128i mmTemp1 = _mm_sub_epi16(mmSrcY1Temp, mmSrcY0Temp);
__m128i mmTempX = _mm_srai_epi16(_mm_add_epi16(mmGradX0, mmGradX1), 3);
@@ -246,8 +246,8 @@ void calcBIOPar_SSE(const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGr
__m128i mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
__m128i mm_h = _mm_unpackhi_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp1 + x), mm_l);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp1 + x + 4), mm_h);
+ _mm_storeu_si128((__m128i *)(dotProductTemp1 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp1 + x + 4), mm_h);
// m_piDotProductTemp2
mm_b = _mm_mulhi_epi16(mmTempX, mmTempY);
@@ -256,8 +256,8 @@ void calcBIOPar_SSE(const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGr
mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
mm_h = _mm_unpackhi_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp2 + x), mm_l);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp2 + x + 4), mm_h);
+ _mm_storeu_si128((__m128i *)(dotProductTemp2 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp2 + x + 4), mm_h);
// m_piDotProductTemp3
mm_b = _mm_mulhi_epi16(mmTempX, mmTemp1);
@@ -266,8 +266,8 @@ void calcBIOPar_SSE(const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGr
mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
mm_h = _mm_unpackhi_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp3 + x), mm_l);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp3 + x + 4), mm_h);
+ _mm_storeu_si128((__m128i *)(dotProductTemp3 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp3 + x + 4), mm_h);
// m_piDotProductTemp5
mm_b = _mm_mulhi_epi16(mmTempY, mmTempY);
@@ -276,8 +276,8 @@ void calcBIOPar_SSE(const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGr
mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
mm_h = _mm_unpackhi_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp5 + x), mm_l);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp5 + x + 4), mm_h);
+ _mm_storeu_si128((__m128i *)(dotProductTemp5 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp5 + x + 4), mm_h);
// m_piDotProductTemp6
mm_b = _mm_mulhi_epi16(mmTempY, mmTemp1);
@@ -286,18 +286,18 @@ void calcBIOPar_SSE(const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGr
mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
mm_h = _mm_unpackhi_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp6 + x), mm_l);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp6 + x + 4), mm_h);
+ _mm_storeu_si128((__m128i *)(dotProductTemp6 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp6 + x + 4), mm_h);
}
- for (; x < ((iWidthG >> 2) << 2); x += 4)
+ for (; x < ((widthG >> 2) << 2); x += 4)
{
- __m128i mmSrcY0Temp = _mm_srai_epi16(_mm_loadl_epi64((__m128i*)(pSrcY0Temp + x)), 6);
- __m128i mmSrcY1Temp = _mm_srai_epi16(_mm_loadl_epi64((__m128i*)(pSrcY1Temp + x)), 6);
- __m128i mmGradX0 = _mm_loadl_epi64((__m128i*)(pGradX0 + x));
- __m128i mmGradX1 = _mm_loadl_epi64((__m128i*)(pGradX1 + x));
- __m128i mmGradY0 = _mm_loadl_epi64((__m128i*)(pGradY0 + x));
- __m128i mmGradY1 = _mm_loadl_epi64((__m128i*)(pGradY1 + x));
+ __m128i mmSrcY0Temp = _mm_srai_epi16(_mm_loadl_epi64((__m128i*)(srcY0Temp + x)), 6);
+ __m128i mmSrcY1Temp = _mm_srai_epi16(_mm_loadl_epi64((__m128i*)(srcY1Temp + x)), 6);
+ __m128i mmGradX0 = _mm_loadl_epi64((__m128i*)(gradX0 + x));
+ __m128i mmGradX1 = _mm_loadl_epi64((__m128i*)(gradX1 + x));
+ __m128i mmGradY0 = _mm_loadl_epi64((__m128i*)(gradY0 + x));
+ __m128i mmGradY1 = _mm_loadl_epi64((__m128i*)(gradY1 + x));
__m128i mmTemp1 = _mm_sub_epi16(mmSrcY1Temp, mmSrcY0Temp);
__m128i mmTempX = _mm_srai_epi16(_mm_add_epi16(mmGradX0, mmGradX1), 3);
@@ -308,78 +308,78 @@ void calcBIOPar_SSE(const Pel* pSrcY0Temp, const Pel* pSrcY1Temp, const Pel* pGr
__m128i mm_a = _mm_mullo_epi16(mmTempX, mmTempX);
__m128i mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp1 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp1 + x), mm_l);
// m_piDotProductTemp2
mm_b = _mm_mulhi_epi16(mmTempX, mmTempY);
mm_a = _mm_mullo_epi16(mmTempX, mmTempY);
mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp2 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp2 + x), mm_l);
// m_piDotProductTemp3
mm_b = _mm_mulhi_epi16(mmTempX, mmTemp1);
mm_a = _mm_mullo_epi16(mmTempX, mmTemp1);
mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp3 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp3 + x), mm_l);
// m_piDotProductTemp5
mm_b = _mm_mulhi_epi16(mmTempY, mmTempY);
mm_a = _mm_mullo_epi16(mmTempY, mmTempY);
mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp5 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp5 + x), mm_l);
// m_piDotProductTemp6
mm_b = _mm_mulhi_epi16(mmTempY, mmTemp1);
mm_a = _mm_mullo_epi16(mmTempY, mmTemp1);
mm_l = _mm_unpacklo_epi16(mm_a, mm_b);
- _mm_storeu_si128((__m128i *)(m_piDotProductTemp6 + x), mm_l);
+ _mm_storeu_si128((__m128i *)(dotProductTemp6 + x), mm_l);
}
- for (; x < iWidthG; x++)
+ for (; x < widthG; x++)
{
- int temp = (pSrcY0Temp[x] >> 6) - (pSrcY1Temp[x] >> 6);
- int tempX = (pGradX0[x] + pGradX1[x]) >> 3;
- int tempY = (pGradY0[x] + pGradY1[x]) >> 3;
- m_piDotProductTemp1[x] = tempX * tempX;
- m_piDotProductTemp2[x] = tempX * tempY;
- m_piDotProductTemp3[x] = -tempX * temp;
- m_piDotProductTemp5[x] = tempY * tempY;
- m_piDotProductTemp6[x] = -tempY * temp;
+ int temp = (srcY0Temp[x] >> 6) - (srcY1Temp[x] >> 6);
+ int tempX = (gradX0[x] + gradX1[x]) >> 3;
+ int tempY = (gradY0[x] + gradY1[x]) >> 3;
+ dotProductTemp1[x] = tempX * tempX;
+ dotProductTemp2[x] = tempX * tempY;
+ dotProductTemp3[x] = -tempX * temp;
+ dotProductTemp5[x] = tempY * tempY;
+ dotProductTemp6[x] = -tempY * temp;
}
- pSrcY0Temp += iSrc0Stride;
- pSrcY1Temp += iSrc1Stride;
- pGradX0 += iGradStride;
- pGradX1 += iGradStride;
- pGradY0 += iGradStride;
- pGradY1 += iGradStride;
- m_piDotProductTemp1 += iWidthG;
- m_piDotProductTemp2 += iWidthG;
- m_piDotProductTemp3 += iWidthG;
- m_piDotProductTemp5 += iWidthG;
- m_piDotProductTemp6 += iWidthG;
+ srcY0Temp += src0Stride;
+ srcY1Temp += src1Stride;
+ gradX0 += gradStride;
+ gradX1 += gradStride;
+ gradY0 += gradStride;
+ gradY1 += gradStride;
+ dotProductTemp1 += widthG;
+ dotProductTemp2 += widthG;
+ dotProductTemp3 += widthG;
+ dotProductTemp5 += widthG;
+ dotProductTemp6 += widthG;
}
}
template< X86_VEXT vext >
void calcBlkGradient_SSE(int sx, int sy, int *arraysGx2, int *arraysGxGy, int *arraysGxdI, int *arraysGy2, int *arraysGydI, int &sGx2, int &sGy2, int &sGxGy, int &sGxdI, int &sGydI, int width, int height, int unitSize)
{
- int *pGx2 = arraysGx2;
- int *pGy2 = arraysGy2;
- int *pGxGy = arraysGxGy;
- int *pGxdI = arraysGxdI;
- int *pGydI = arraysGydI;
+ int *Gx2 = arraysGx2;
+ int *Gy2 = arraysGy2;
+ int *GxGy = arraysGxGy;
+ int *GxdI = arraysGxdI;
+ int *GydI = arraysGydI;
// set to the above row due to JVET_K0485_BIO_EXTEND_SIZE
- pGx2 -= (JVET_L0256_BIO_EXTEND_SIZE*width);
- pGy2 -= (JVET_L0256_BIO_EXTEND_SIZE*width);
- pGxGy -= (JVET_L0256_BIO_EXTEND_SIZE*width);
- pGxdI -= (JVET_L0256_BIO_EXTEND_SIZE*width);
- pGydI -= (JVET_L0256_BIO_EXTEND_SIZE*width);
+ Gx2 -= (BIO_EXTEND_SIZE*width);
+ Gy2 -= (BIO_EXTEND_SIZE*width);
+ GxGy -= (BIO_EXTEND_SIZE*width);
+ GxdI -= (BIO_EXTEND_SIZE*width);
+ GydI -= (BIO_EXTEND_SIZE*width);
__m128i vzero = _mm_setzero_si128();
__m128i mmGx2Total = _mm_setzero_si128();
@@ -388,13 +388,13 @@ void calcBlkGradient_SSE(int sx, int sy, int *arraysGx2, int *arraysGxGy
__m128i mmGxdITotal = _mm_setzero_si128();
__m128i mmGydITotal = _mm_setzero_si128();
- for (int y = -JVET_L0256_BIO_EXTEND_SIZE; y < unitSize + JVET_L0256_BIO_EXTEND_SIZE; y++)
+ for (int y = -BIO_EXTEND_SIZE; y < unitSize + BIO_EXTEND_SIZE; y++)
{
- __m128i mmsGx2 = _mm_loadu_si128((__m128i*)(pGx2 - 1)); __m128i mmsGx2Sec = _mm_loadl_epi64((__m128i*)(pGx2 + 3));
- __m128i mmsGy2 = _mm_loadu_si128((__m128i*)(pGy2 - 1)); __m128i mmsGy2Sec = _mm_loadl_epi64((__m128i*)(pGy2 + 3));
- __m128i mmsGxGy = _mm_loadu_si128((__m128i*)(pGxGy - 1)); __m128i mmsGxGySec = _mm_loadl_epi64((__m128i*)(pGxGy + 3));
- __m128i mmsGxdI = _mm_loadu_si128((__m128i*)(pGxdI - 1)); __m128i mmsGxdISec = _mm_loadl_epi64((__m128i*)(pGxdI + 3));
- __m128i mmsGydI = _mm_loadu_si128((__m128i*)(pGydI - 1)); __m128i mmsGydISec = _mm_loadl_epi64((__m128i*)(pGydI + 3));
+ __m128i mmsGx2 = _mm_loadu_si128((__m128i*)(Gx2 - 1)); __m128i mmsGx2Sec = _mm_loadl_epi64((__m128i*)(Gx2 + 3));
+ __m128i mmsGy2 = _mm_loadu_si128((__m128i*)(Gy2 - 1)); __m128i mmsGy2Sec = _mm_loadl_epi64((__m128i*)(Gy2 + 3));
+ __m128i mmsGxGy = _mm_loadu_si128((__m128i*)(GxGy - 1)); __m128i mmsGxGySec = _mm_loadl_epi64((__m128i*)(GxGy + 3));
+ __m128i mmsGxdI = _mm_loadu_si128((__m128i*)(GxdI - 1)); __m128i mmsGxdISec = _mm_loadl_epi64((__m128i*)(GxdI + 3));
+ __m128i mmsGydI = _mm_loadu_si128((__m128i*)(GydI - 1)); __m128i mmsGydISec = _mm_loadl_epi64((__m128i*)(GydI + 3));
mmsGx2 = _mm_add_epi32(mmsGx2, mmsGx2Sec);
mmsGy2 = _mm_add_epi32(mmsGy2, mmsGy2Sec);
@@ -409,11 +409,11 @@ void calcBlkGradient_SSE(int sx, int sy, int *arraysGx2, int *arraysGxGy
mmGxdITotal = _mm_add_epi32(mmGxdITotal, mmsGxdI);
mmGydITotal = _mm_add_epi32(mmGydITotal, mmsGydI);
- pGx2 += width;
- pGy2 += width;
- pGxGy += width;
- pGxdI += width;
- pGydI += width;
+ Gx2 += width;
+ Gy2 += width;
+ GxGy += width;
+ GxdI += width;
+ GydI += width;
}
mmGx2Total = _mm_hadd_epi32(_mm_hadd_epi32(mmGx2Total, vzero), vzero);
@@ -798,7 +798,7 @@ void PelBufferOps::_initPelBufOpsX86()
addAvg8 = addAvg_SSE<vext, 8>;
addAvg4 = addAvg_SSE<vext, 4>;
-#if JVET_L0256_BIO
+#if ENABLE_SIMD_OPT_BIO
addBIOAvg4 = addBIOAvg4_SSE<vext>;
bioGradFilter = gradFilter_SSE<vext>;
calcBIOPar = calcBIOPar_SSE<vext>;
diff --git a/source/Lib/CommonLib/x86/RdCostX86.h b/source/Lib/CommonLib/x86/RdCostX86.h
index ab54737fa5a9e3e47c64660d9f4ab73d06624827..4d87189ca052e0d8ab14a17980b92c2e1406c13e 100644
--- a/source/Lib/CommonLib/x86/RdCostX86.h
+++ b/source/Lib/CommonLib/x86/RdCostX86.h
@@ -297,7 +297,7 @@ Distortion RdCost::xGetSAD_SIMD( const DistParam &rcDtParam )
return uiSum >> DISTORTION_PRECISION_ADJUSTMENT(rcDtParam.bitDepth);
}
-#if JVET_L0256_BIO
+#if ENABLE_SIMD_OPT_BIO
template< X86_VEXT vext >
Distortion RdCost::xGetSAD_IBD_SIMD(const DistParam &rcDtParam)
{
@@ -308,14 +308,14 @@ Distortion RdCost::xGetSAD_IBD_SIMD(const DistParam &rcDtParam)
const short* src1 = (const short*)rcDtParam.cur.buf;
int width = rcDtParam.org.height;
int height = rcDtParam.org.width;
- int iSubShift = rcDtParam.subShift;
- int iSubStep = (1 << iSubShift);
- const int src0Stride = rcDtParam.org.stride * iSubStep;
- const int src1Stride = rcDtParam.cur.stride * iSubStep;
+ int subShift = rcDtParam.subShift;
+ int subStep = (1 << subShift);
+ const int src0Stride = rcDtParam.org.stride * subStep;
+ const int src1Stride = rcDtParam.cur.stride * subStep;
__m128i vtotalsum32 = _mm_setzero_si128();
__m128i vzero = _mm_setzero_si128();
- for (int y = 0; y < height; y += iSubStep)
+ for (int y = 0; y < height; y += subStep)
{
for (int x = 0; x < width; x += 4)
{
@@ -332,7 +332,7 @@ Distortion RdCost::xGetSAD_IBD_SIMD(const DistParam &rcDtParam)
vtotalsum32 = _mm_hadd_epi32(vtotalsum32, vzero);
Distortion uiSum = _mm_cvtsi128_si32(vtotalsum32);
- uiSum <<= iSubShift;
+ uiSum <<= subShift;
return uiSum >> DISTORTION_PRECISION_ADJUSTMENT(rcDtParam.bitDepth);
}
#endif
@@ -2462,7 +2462,7 @@ void RdCost::_initRdCostX86()
m_afpDistortFunc[DF_HAD64] = RdCost::xGetHADs_SIMD<Pel, Pel, vext>;
m_afpDistortFunc[DF_HAD16N] = RdCost::xGetHADs_SIMD<Pel, Pel, vext>;
-#if JVET_L0256_BIO
+#if ENABLE_SIMD_OPT_BIO
m_afpDistortFunc[DF_SAD_INTERMEDIATE_BITDEPTH] = RdCost::xGetSAD_IBD_SIMD<vext>;
#endif
}